The Labubu craze: Using NLP to understand the behaviour behind market surges and bubbles

How can behavioural data help us spot rising and disappearing bubbles? The recent surge in popularity of Labubu dolls led to rising prices for both the toys themselves and for the share prices of their Chinese manufacturer, Pop Mart. Ideally investors want to identify a bubble forming, and more importantly when it might burst, to maximise the potential to outperform. In the past, accomplishing this has been difficult because bubbles are driven by human behaviour, which has been difficult to quantify. Now, it’s possible to use artificial intelligence (AI) – notably natural language processing (NLP) – to turn unstructured text from news stories, social media, and more into structured data for statistical processing to generate behavioural insights. This can enable traders and investors to spot and track bubbles across markets. In this insight we discuss:

• The surge in the popularity of the Labubu dolls as well as their manufacturer’s share price has led to a potential bubble for both.
• Predicting the build-up and bursting of bubbles has been challenging until now because it requires forecasting collective buying and selling behaviour, which is notoriously difficult.
• By leveraging NLP, market participants can monitor news and social media to determine when euphoric demand for a consumer good shows signs of waning.

The graphic above shows a count of Labubu references across thousands of global financial news & social media sources, with the Pop Mart International (9992.HK) stock price superimposed. What’s clear now is that the attention and perception towards these dolls is tightly linked to Pop Mart’s success, and changes in public sentiment have provided important markers of the novelty potentially wearing off.

The small, extremely popular Labubu plush dolls with fierce grins have seen their prices surge, partly thanks to celebrities like Kim Kardashian and Lisa from K-pop group Blackpink endorsing them on social media. Now, prices for the rarer Labubus are more than $1,000 each. The question is, have we hit peak Labubu? Is it possible to predict the bursting of a Labubu bubble, if there currently is one?

Ideally investors could learn from previous bubbles how to manoeuvre their portfolios through new ones. Historical examples of bubbles include the dot-com era, during which internet stock valuations surged, and then collapsed. Parallel to Labubu, Beanie Babies (another group of small plush toys) saw a bubble form in the late 1990s. Some Beanie Baby versions, which would have retailed for $5, climbed to as high as $500, before the market in them dried up.

Being able to predict the peak of a bubble and sell the assets concerned before a precipitous fall would be ideal for investors. But predicting when a bubble will burst is a tricky business, because bubbles are behavioural phenomena that have been hard to model quantitatively.

Sales of Labubu dolls in America were up by more than 1,200% in the three months ending September 2025, and in Europe they had climbed over 700% [Note 1]. Pop Mart, the Chinese manufacturer of these dolls, saw its global revenue jump 250% during the same period. Pop Mart shares also rose strongly during the third quarter, giving it a stock market value of about £34 billion. While the secondary market for the Labubu dolls themselves cooled in the early Q4 2025, we can build a clearer picture of what peak Labubu looks like by turning our attention to news & social media.

With AI (specifically NLP), investors can understand the sentiment within what is written about Labubu. This largely unstructured data can be processed into structured insight, which investors then have the potential to translate into alpha.   

NLP is a specialised branch of AI that enables computers to interpret, process, and analyse human language at scale. NLP encompasses various operations on text, including recognising and classifying entities (e.g., companies, products, people), analysing sentiments (positive, neutral, negative) and emotions (e.g., excited, optimistic, annoyed), and extracting events from unstructured text. NLP is able to turn large text datasets into standardised labels and metrics useful for statistical processing and research. This data can then be analysed on its own or in relationship with other data, such as equity prices.

LSEG MarketPsych NLP Engine creates structured datasets with precision tagging from unstructured text (such as Reuters news, social media, transcripts, a client’s proprietary content and more). The NLP-as-a-service platform identifies key analytics from each sentence, including:

• Entities: Millions of entities are identified and tagged from one of over 20 categories.
• Topics: Over 1,000 categorised topics are labelled (such as bankruptcies, clinical trials, and acquisitions)
• Sentiments & Emotions: For each sentence, financial, ESG, and commodities sentiment are classified. Fourteen emotional tones are measured in first-person commentary.
• Events: Over 4,000 events, with dependency labels for context, are tagged.

In summary, the resulting behavioural insights can be powerful. Identifying Labubu mentions in both news and social media, the NLP Engine’s analytics provides statistical inputs to help investors spot the development of a bubble, and when the bubble might be about to burst. For example, it could help a trader to spot the waning of the dolls’ popularity, and therefore a decline in Pop Mart’s share price.

LSEG MarketPsych NLP Engine can be deployed via API. In addition, LSEG MarketPsych provides hosting advice and retrieval and visualisation guidance for the detailed analytics. Contact us to explore how LSEG MarketPsych NLP Engine can be integrated into workflow.